Watch disc A move toward disc B. When they meet, something happens — or seems to happen. Adjust the delay between contact and B's departure. Notice where the percept flips.
In 1946 the Belgian psychologist Albert Michotte published La Perception de la Causalité. He showed that causation, like color or motion, can be a direct perceptual primitive — not a conclusion drawn from watching sequences, but something the visual system computes automatically, below the level of deliberate inference.
The timing threshold is the key finding. Below roughly 100–150 ms between contact and B's departure, observers reliably report launching. Above that window, the causal percept breaks: the events become two separate things that happened to coincide. This is not a belief you can override. Knowing the display is artificial doesn't help. The computation runs and delivers its verdict before deliberate reasoning gets involved.
Michotte identified several variants. In entraining, A and B move together after contact as though A is pushing B. In triggering, B's velocity after contact greatly exceeds A's — A seems to have released something rather than transferred momentum. The parameter space maps onto an informal folk physics: launching, triggering, pulling, carrying.
David Hume argued in 1748 that causality is never perceived — only regular succession, from which the mind infers cause. Michotte's experiments look like a direct counter-example. The philosophical response was to say: this is still just inference, only very fast. But that raises a question about what "perception" is supposed to mean. We don't perceive light wavelengths, we perceive colors. We don't perceive air pressure oscillations, we perceive voices. All perception involves computation. The claim that causal perception is "mere inference" requires a principled reason why this computation doesn't count.